Kara Santamaria

Part of it is designing your experiment, coming up with all the protocols, figuring out how you're going to measure everything, and then how you're going to, you know, iterate as you go.

And the other part is the design.

dexterity component it's it's actually being good at doing the physical lab work often you have lab techs who are involved in that a lot of graduate students and undergraduate students but that part is more and more we are capable of completely taking that part over with machines that doesn't feel risky to most people right it actually feels like a boon okay so people introduce error and

But a robot can do it faster, more efficiently, with less waste, you know, with less contamination.

That all sounds great.

But then what happens when the design part of the experiment starts to fall into the hands of A.I.

?

Now we're applying all of those same improvements, faster, more efficient, cleaner, with less waste, to a protocol that used to take hours to days to months to years, now can happen in seconds to minutes.

And so what does that mean for biosecurity?

And this is a real concern.

So they call this the dual use problem.

The same AI tools that could cause great benefit to an entire people can also be misused and cause real harm.

They can be repurposed, but they can also be very intentionally used in that way.

So, we've got a few studies already that have been published and are available about ways that this has been done.

So, we're not just talking, you know, pontificating and saying, hmm, what could happen?

But a few studies where they're like, okay, let's see if this actually can and has happened.

So I've got a study here I can cite where protein evolution was enhanced using large language models and they were able to improve enzyme activity, mutation rate.

and evolution significantly from traditional methods.

We've got a study here where viral spread was optimized.

So that's fun, right?